First, the prediction targets do not seem to be picked randomly. Using a Senegenin heatmap of the gene expression data, it can be easily seen that the prediction target genes are highly perturbed by 3AT treatment. Second, the three knockout genes are transcription factors, and their binding targets can be obtained from ChIP-chip data. Finally, given the relatively large number of available data points and the small number of target genes, the problem size seems to be reasonable to be handled by the existing network construction algorithms. However, I decided not to pursue gene regulatory networks for this problem, for reasons stated above regarding to inter-data set consistency, and also because most network reconstruction algorithms are model-driven, relying on Sesamoside simplifying model assumptions that are often hard to be tested or fulfilled. For example, methods for constructing regulatory networks must make some simplifying assumptions that may not be true. For example, most methods assume that the mRNA level of a regulator is a true indication of its activity, and that there is no time lag or a constant time lag between the transcription of a regulator and the transcription of its target genes. In reality, some regulators may be regulated post-transcriptionally or on the protein level, with no change on their transcriptional levels. Also, transcriptional time lags between regulators and target genes are not constant and are difficult to estimate in general. In contrast, the simple co-expressionbased methods that I have taken assume that gene expression levels in the prediction strain are somewhat correlated with that in the other strains, an assumption that can be easily tested. It would be very interesting to know what methods the other participants have used, especially the methods that have had inferior performance. Unfortunately, except for the GH method that shared ����top performer���� status with KNN, details of the other methods are not disclosed, making it hard to speculate why the other methods did not work well. Given that the main theme of the challenge is to evaluate reverse-engineering methods, I suspect some participates have attempted to construct gene regulatory networks.